One aim of this research proposal is first to empirically establish an ordering with respect to the difficulty of learning various functional relationships between multiple cues and their criterion. A critical feature of the design of the proposed studies is the use of extrapolation tests that require subjects to make predictions using new cue values that go beyond, or lie outside, the range of training experience. Performance on these tasks will provide a much needed empirical data base for testing the rule versus associative models of function learning. The ultimate goal of this research proposal is to develop a mathematical model of multiple cue function leaning in a common framework with current models of category learning.